Claude's new Fable model dropped, so naturally the first thing I built was a Sasquatch

Everyone's first move with a new frontier model is the same: scaffold a SaaS, wire up Stripe, benchmark it against their backlog. When Anthropic's Fable became available, I gave it a different kind of test — the highest-value project I could think of. A campfire.

squatchy is a Pacific Northwest scene rendered entirely in ASCII art, in the spirit of the classic asciiquarium: two ridgelines of snow-capped mountains, the Sound below, evergreen forest, a crackling campfire — and wildlife wandering through. Deer, elk, bears, eagles, leaping salmon, and orcas that surface and blow like the real thing. Day fades into night, stars come out, rain rolls in on its own (it's the PNW, after all). And if you're patient, the big guy himself ambles past the treeline.

One Python file. ~1,400 lines. Zero dependencies. Runs in any terminal on macOS or Linux.

GitHub: https://github.com/HammaRidgeVentures/squatchy

Here's why this was the right test. After 30 years of shipping web apps, I don't write code line by line anymore — I build ideas. And a CRUD app tells you nothing about a model; there are ten thousand of those in the training data. But "make an orca porpoise convincingly in ASCII, in curses, at terminal frame rates"? No spec for that. No Stack Overflow answer. Just taste, iteration, and me saying "the spout comes before the roll" until it felt like the real thing off Puget Sound.

That's the whole review, honestly: the orca looks right. The fireflies drift out of the treeline at dusk. The big guy shows up when he feels like it.

When building costs almost nothing, the scarce skill is having ideas worth building — and finishing them. So here's my question: new model drops, you get one evening with it. Do you test it on your roadmap, or on something gloriously pointless?